The numbers arrive with the mute finality of a ledger entry. In the first quarter of 2025, 57 percent of all venture capital deployed in the technology sector went to artificial intelligence companies. Not fintech. Not biotech. Not crypto. AI. The concentration is not a trend; it is a tectonic shift. And beneath the glossy headlines announcing billion-dollar rounds for model labs, the venture capital ecosystem is fracturing along a fault line that was always there but has only now become unmissable: the gap between funds that can write nine-figure checks and funds that cannot.
I have watched this dissassembly from a particular perch. Vienna is not San Francisco, and my work as a due diligence analyst has never been about chasing the herding instinct of Sand Hill Road. It is about measuring the distance between what a project claims and what its architecture will actually support. Over the past nine months, I have audited eleven venture funds, four of them below $200 million in assets under management. The pattern in their deal flow is not subtle. They are no longer trying to compete for AI's foundational layer. They are running toward the periphery, hoping that the gravitational pull of the frontier model labs does not drag them into irrelevance.
This piece is not an obituary for small funds. It is an x-ray of the structural forces now determining who gets to participate in the most consequential technology cycle since the internet. The code does not lie, but the contract can. And the contracts being written in venture capital right now are worth reading carefully.
The venture capital industry was designed for a world of atomized risk. Diversify across thirty early-stage companies, three of whom will return the fund, and the math works. The model made sense when a $50 million fund of funds could place $10 million into ten seed. It made less sense as round sizes ballooned to match the capital intensity of AI development. The model breaks entirely when a single company requires more capital than your entire fund has managed in a decade.
Rise of the mega-round is a structural bet, not a judgment call.
Silence is the loudest indicator of risk. When I sit in partner meetings with small funds, the conversation around AI has become quieter. Not because they intend to avoid it, but because they know that competing for allocation in a $6 billion OpenAI round, which a16z and Thrive Capital can absorb without pain, is a fantasy for a fund whose total AUM is a rounding error in that same term sheet. What emerges is a quiet acknowledgment that the new economy of venture rewards scale so dramatically that the old playbook no longer applies. But what does that mean for the actual construction of portfolios, the sourcing of deals, and the ability to provide real support to companies? That is where the story is.
Aesthetics matter in venture capital, just as they do in crypto. A well-constructed portfolio deck, clean positioning, quarterly reporting that shows disciplined follow-on reserves. It is the mask. The geometry beneath is simpler and harsher: entry price, ownership percentage, expected exit multiple. When the entry price at the seed stage for an AI company is $50 million post-money and the company will need $200 million more before it generates substantial revenue, the geometry of a small fund's return profile collapses. Even if the company succeeds, the dilution means a small fund's ownership is compressed into an outcome too small to move a $100 million fund's net return. The trade is not worth the risk. And so they move on to projects that math can accommodate.
The venture capital industry has always been a barbell. At one end, you have the early stage: high risk, high uncertainty, high optionality. At the other, you have the late stage: lower risk, higher liquidity, but lower headline return. The middle has always been the safest place to be. The AI wave has hollowed out that middle. A $400 million fund once had clear competition in Series A and B rounds for software companies that needed $20 million to grow into $200 million revenue. Today, those same rounds are being replaced by AI infrastructure companies raising $100 million just to test a model, or application companies raising $30 million before they have even signed their first enterprise contract. The check size has to scale with the ambition of the pitch. If the pitch is based on model capability that demands massive compute, the capital requirement is not decoupled from the technical requirement. It is intrinsic to it.
I have seen this dynamic play out in the crypto industry in ways that many observers still misread. The honest truth is that AI is not competing with crypto for narrative supremacy. It is competing for the same LP allocation. When a pension fund's venture allocation is fixed, a dollar into an AI mega-round is a dollar withdrawn from the pool that might have gone into a crypto fund, a fintech fund, or a consumer fund. The result is an industry-wide liquidity squeeze for the non-AI sectors, and this is not a cyclical phenomenon. It is a portfolio reallocation, like a storage drive being erased to make room for new data.
We are told by market optimists that small funds will simply pivot. They will move into AI applications, find niche verticals, and create value through domain expertise rather than financial scale. This is half-true, and that half-truth is dangerous. The part that is true is that capital is now necessary for AI model development at a scale that demands consolidation. The part that is false is the assumption that application-level AI requires no specialized infrastructure expertise or that small funds can provide it with their traditional toolkits. If you are a $150 million fund and you are evaluating an AI-driven legal research company, you must now understand whether the underlying model is fine-tuned on licensed court data or on pirated transcripts. You must know whether the latency of the API is acceptable, whether the inference cost at scale is survivable, and whether the market's willingness to pay is based on real savings or on hype. That due diligence skill set is different from what most small funds deployed for the last decade, when they were evaluating API wrappers over Stripe or chat widgets for customer service.
One emerging counterargument suggests that small funds can still find deal flow that the large funds overlook. In some ways, that is true. The crypto winter of 2022, for example, created a window for disciplined funds to acquire positions at favorable prices when the megafunds were still licking their wounds from FTX and Terra. During that same winter, I advised several funds to concentrate on infrastructure opportunities and to ignore the consumer narratives that were dominating the press cycle. The funds that listened exited 2023 with portfolios that looked meaningfully different from the broader market, focused on technical infrastructure rather than consumer hero narratives. That divergence was not genius. It was the simple discipline of looking at the yield curve of innovation and being patient enough to wait for the right entry.
But that kind of discipline is becoming harder to practice. The innovation curve is steeper, the capital requirements are larger, and the timing of exits is longer. "I do not follow the wave; I measure its depth" has become a survival mantra for the funds I work with, and depth measurement has shown increasingly alarming signs.
Consider the specific mathematics of an AI investment for a $150 million venture fund. If they take an 8 percent position in a seed round that leads to a $30 million valuation, they invest $2.4 million. For that position to contribute meaningfully to their fund returns, they need a realization of at least $15 million at exit, a 6.25x multiple. That is achievable if the company becomes a $500 million exit and their position is preserved. But AI companies raise more, and dilution is sharper. If the company raises three additional rounds at increasing valuations, and the fund does not have the reserves to maintain its ownership percentage, it will be diluted from 8 percent to 3 percent. At that point, the same $500 million exit only returns $15 million before taxes, which reverts to $7.5 million after carry. For a $150 million fund, that result is less than the cost of a full-time position over a decade. The trade fails. The moment that becomes clear, the small fund's AI allocation decision turns from a technical analysis into an existential one.
This is the hidden structure beneath the seemingly simple observation that smaller funds are "being left behind." They are not being left behind as observers. They are being structurally excluded by the geometry of fund accounting. The optimum portfolio size shifts as the cost of participation changes. When the capital required to participate in a technology wave exceeds a fund's total AUM, the wave becomes mathematically unavailable, and the only rational strategy is to find a parallel ecosystem or to wait for the next cycle entirely.
Now, we have a growing category of "crossover" funds that are beginning to look more like private equity operations than classic venture. They write checks of $100 million or more, take board seats with real influence, and provide strategic support that goes beyond capital. This is not a new trend in venture. But what is new is the extent to which the top AI companies have demanded this type of investor. The narrative around OpenAI, Anthropic, and other frontier labs has shifted. They are no longer looking for merely financial backers. They are looking for capital partners with cloud credits, GPU access, geopolitical connections, and the ability to navigate regulatory hurdles across borders. These are not capabilities that a $200 million fund can offer, even if the partners have a decade of relevant experience.
The impact of this divergence is being felt in measurable ways across the ecosystem. Deal volume at the seed and Series A stages for traditional SaaS companies is down, on a relative basis, by roughly 30 percent. The share of new venture funds being launched by first-time GPs is also declining, as LP enthusiasm for new managers wavers in the face of the AI mega-rounds. When the limited partners write their allocation letters for the next calendar year, they are increasingly sending a simple signal: AI exposure matters more than manager freshness.
For the crypto industry specifically, this is a moment of profound reallocation. I have spent the last several years writing that "beneath the yield lies the rot" when it comes to DeFi protocols that promise returns without sustainable value generation. The AI capital concentration is a different kind of rot. It is not a failure of a specific project, but a systemic drain of attention, capital, and talent that is pulling the substrate out from under alternative sectors. If you are a venture fund that focused exclusively on digital assets, your LP outreach has become significantly more difficult. You are now competing with funds that have a "crypto plus AI" or "AI plus crypto" narrative, and the central term in that formula is increasingly the AI component. The crypto element is becoming a differentiator, not the core, and that is a fundamental shift.
Let me be precise about what I am not saying. I am not saying that AI is a bubble that will soon pop, sending all capital back to crypto. That narrative, popular among some crypto maximalists, is a deflection. AI's economic fundamentals are real, and the commitment to compute is not merely a form of speculative mania. But I am saying that the new capital cycle has structural implications for venture capital that will outlast any temporary corrections in AI valuations. The venture industry's evolution toward consolidation is as old as its history. In 1999, the top ten venture funds deployed about 22 percent of all capital. By 2004, that number had risen to 35 percent. By 2014, it had passed 50 percent. And today, the top ten funds manage approximately half of the industry's total assets. AI has accelerated this trend, not created it. The direction was already set. AI simply added a jet engine to the existing trajectory.
What happens to the 2,600 small venture funds currently operating in the United States is the real question. They are the lifeblood of early-stage innovation. They fund the unglamorous B2B startups, the university spinouts, the first-time founders who will build the next decade's companies. If they are systematically squeezed out of the transformative technology cycle, the entire ecosystem loses diversity. The industry's innovation engine, so dependent on tinkering and iteration, risks becoming a monoculture funded by a small group of megafunds that all share the same worldview, the same access to the same deals, and the same biases.
A monoculture in any environmental system is a sign of fragility. A venture monoculture is a similar signal. The ecosystem that produced WhatsApp, Stripe, and Ethereum was built on a wide base of small and medium funds willing to bet on individual founders at an early stage. If the AI concentration leads those funds to abandon their core competency in search of AI exposure, everyone loses. The innovation will be less diverse, the founder experience will be more homogenous, and the market will be less resilient to sector-specific shocks.
This is not a purely theoretical concern. I have observed concrete examples of the behavior change. The fund I work with in Vienna has a mandate that is technology-neutral, but the practical realization of that mandate has shifted. Two years ago, we were evaluating blockchain interoperability protocols and privacy infrastructure. Today, the evaluation queue is dominated by AI reliability infrastructure, AI data sovereignty, and AI for regulatory compliance. The shift in deal flow mirrors the shift in LP conversations. When we speak to our own limited partners, the first question is no longer "what is your crypto thesis?" It is "how are you positioned for the AI transition?" And the only honest answer we can give is that we are positioned the way a small fund can be: selective, focused, and willing to be patient. That is not a comforting answer. It is a realistic one.
The landscape is not void of opportunities for the smaller funds, however. There are countercurrents. The AI trends have created secondary opportunities in security, data provenance, and verification tooling that require specialized expertise and are less attractive to megafunds. The cryptocurrency industry has a natural intersection with these requirements, particularly in areas such as decentralized computation, verifiable inference, and proof-of-training protocols. These are not the shiny consumer apps that get headlines, but they are the infrastructure layers that will be needed as AI integrates into regulated industries like healthcare and finance. A small fund that genuinely understands both the technical architecture of these systems and the regulatory landscape in which they will operate can still find an edge that a megafund cannot easily replicate.
One such field, which I have studied closely, is the emerging area of "AI compliance infrastructure" — tools that allow enterprises to prove that their model outputs are reproducible, their training data is appropriately licensed, and their inference costs are auditable. These requirements are not naturally aligned with the AI companies' own incentives. But they are becoming non-negotiable for institutions that plan to deploy AI in high-stakes contexts. The code does not lie, but the contract can. This is where small funds can build a constructive bridge between the technical frontier and the compliance reality.
Yet there is a tension in this strategic pivot. The same small funds that are moving toward AI-adjacent infrastructure are still being asked to show alpha in their core fund being. If their LPs wanted AI infrastructure exposure, they could buy the public equities of Nvidia, Microsoft, or a cloud provider. What a small fund must provide is superior capital deployment and decision-making within a specialized niche, not merely a correlated asset allocation. That requires a level of focus that is difficult to maintain when the market narrative pulls in a different direction. The discipline to say no, to stay within your circle of competence, is perhaps the most undervalued skill in a market driven by FOMO.
As a diagnostician, I am more interested in the structural mechanics of this shift than in which funds will individually survive. What matters is the condition of the broader system. The capital concentration itself is not necessarily bad. It reflects the increasing capital intensity of the technology industry, and that is a rational response to the cost of compute and the scale of data required to build frontier models. But the consequences of that concentration are mixed. On the one hand, it enables breakthroughs that would be impossible for a fragmented cohort of small funds to finance. On the other hand, it concentrates decision-making power in a small number of individuals, who, however capable, are not omniscient. A system in which a handful of funds determine which technology reaches the market is a system that will miss the wilder, more speculative opportunities. It will also be more vulnerable to correction when a consensus thesis fails.
For crypto specifically, the long-term implication is severe. If large AI bets continue to dominate venture capital, the available pool of institutional capital for crypto projects will continue to shrink. Venture funds that were once dedicated to digital assets will either pivot or struggle. The crypto industry will have to rely more heavily on internal sources of capital, such as treasuries from successful protocols, and on retail participation. That is a fundamentally different ecosystem than the one many founders and investors entered in 2017 or 2020. The shift will force a more disciplined focus on revenue generation and actual usage rather than speculative token price appreciation.
I have made a career of finding the structural flaws in projects that others find compelling. The best argument in favor of this current AI consolidation is that it is simply the market's response to the underlying economics. Training runs for frontier models now cost in the range of $100 million or more. Inference costs for large-scale deployments are ongoing expenses that require predictable revenue or massive war chests. No one would expect a small fund to participate in financing the construction of a new dam or a semiconductor fabrication plant. AI model development has become similarly capital-intensive. In this view, the "division" in venture capital is not a failure or a bug. It is a rational reallocation of resources toward the most productive use of capital, and smaller funds should adapt or face extinction.
That argument is not wrong. The problem is that it assumes that all AI value creation lives at the frontier, which is an assumption worth challenging. The actual value in AI will be distributed across an enormous range of applications, many of which do not require frontier-class models. An AI-powered contract review tool for small law firms does not need a trillion-parameter model. It needs a well-designed product, a deep understanding of legal workflows, and a distribution channel. These characteristics are closer to what a traditional venture fund knows how to evaluate. This is where small funds can still participate. They cannot be the lead investor in the infra layer for a frontier lab, but they can be the first institutional check in a vertical AI application that generates revenue within twelve months. Their discipline, their hard-earned domain expertise, and their ability to nurture a company through its early operating years can make them the better investor for that particular company.
This is the "Contrarian angle" that the market narrative misses. The megafunds are not the only winners in the AI cycle. The companies that build AI applications for specific industries — healthcare, legal, manufacturing, government — may not need the massive capital raises of the foundation labs. Their capital efficiency is higher, their path to revenue is clearer, and their competitive advantage is not dependent on an unassailable compute advantage but on their ability to integrate AI into complex human workflows. A small fund that specializes in a vertical industry and understands that industry's data landscape has a genuine edge.
I think of the healthcare AI company that I audited last quarter. It was not the kind of company that would get headlines. No one from the top ten venture funds had bothered to look at it because its total addressable market within its first three years was estimated at only $400 million — too small to matter for a $10 billion fund, but enormous in the context of a $120 million fund's return targets. The company had developed an AI system to pre-authorize insurance claims, cutting the process from weeks to minutes. It was not a transformative technology in the way that a frontier model is. But it was a tool that saved insurers millions of dollars per year, had a clear regulatory pathway, and could be sold easily to a strategic acquirer. The company's technical architecture was elegant, and its unit economics were strong. This is where the small fund can still compete. The code works, the customer is real, and the economics close.
This is the bridge between "big AI aesthetics" and "small fund discipline." The frontier labs raise billions to build the model. The small funds find the application, the distribution channel, and the customer willingness to pay. They perform the work of adoption, which is as important as innovation. The megafund can write a $100 million check, but they cannot make a hospital trust an AI output with their name on it. The trust is built on patient interaction, prompt resolution of errors, and regulatory compliance.
That is where the venture ecosystem's diversity matters. It is not just about LP returns. It is about ensuring that the technology is applied to problems beyond the narrow universe that captures the imagination of large capital. It is about geographic diversity, sectoral diversity, and founder diversity. The market is not naturally wired to preserve those values. In fact, the market is wired to concentrate and homogenize. The current AI wave accelerates that consolidation. If we do not push back, we will end up with a venture ecosystem that looks more like an oligarchy and less like an ecosystem.
There is, however, a contrary argument from the other side. Some observers point out that the AI mega-rounds are not necessarily coming at the expense of smaller funds in absolute terms. They note that the total venture capital industry is still allocating large sums to non-AI sectors, and the smaller funds' challenge is not a lack of capital but a lack of the new technical literacy required to evaluate AI deals. Under this view, the concentration is a short-term imbalance that will correct as open-source models commoditize the technology. In time, they argue, AI will become just another software component, and the capital requirements will fall. The frontier labs will consolidate, but a new layer of efficient, small-scale companies will emerge. That argument has merit, but it underestimates the time it will take for the commoditization to occur, and during that time, the venture industry's structure will harden.
The precedent from crypto is instructive. The 2021 boom saw a huge influx of capital into digital asset funds, with many new entrants promising exceptional returns. The 2022 collapse wiped out a large portion of those funds. The survivors are the ones with real technical expertise and disciplined risk management. The same process is now happening in AI. The capital concentration is a sign of a market that has not yet separated the wheat from the chaff. When the correction comes, as it will, the market will reveal which AI companies are actually creating value and which were merely narrative constructs.
The question is what happens to the small funds during this interim period. Some will be forced out entirely, and their teams will join larger institutions. Some will pivot and become "AI infrastructure funds" without any real differentiation. A few will find the path that lies between pure frontier exposure and vertical application, the path of AI verification, data provenance, or model security. Those are the funds that will thrive. And their survival is not just a matter of their own returns. It is a matter of maintaining the ecosystem's heterogeneity.
I am often asked, in the context of due diligence, whether a particular project is "safe." The question is usually grounded in an attempt to predict a market outcome. My answer is different. Safety is a function of structure, not prediction. A small fund that has deep domain knowledge, a strong network in a specific industry, and a portfolio that is deliberately constructed to avoid catastrophic losses will be safe in the sense that it can survive any single investment failing. A small fund that chases AI narratives without a genuine understanding of the technology is not safe, regardless of how good the market environment is.
This is the basic lesson of the crypto winter and one that applies equally to the AI boom. The market is not a judge; it is a mirror. It reflects the quality of the underlying capital deployment. When the deployment is based on genuine structural insight, the returns follow. When it is based on narratives and aesthetics, the returns revert to the mean, which is to say, they become inadequate for the risk taken. Hype is noise; structure is signal. The funds that understand this will be the ones who navigate the AI transition without being devoured.
So what should a small fund do in the face of this AI concentration? The first step is an honest assessment of whether their founding thesis is compatible with the new capital cycle. If the fund was built to invest in consumer applications, the AI impact will be severe. Consumer apps are increasingly powered by AI, and a fund that lacks the technical ability to evaluate the AI component will be at a permanent disadvantage. The second step is to recognize that the time for "hedging" is over. A small fund cannot be half-in, half-out. A strategic direction that says "we invest in AI for the enterprise, but our focus is on data quality and verification" is a differentiated position. A direction that says "we invest in AI, cleantech, fintech, and consumer" is a direction that lacks conviction and will fail when the market becomes more difficult.
The final step, and the one that is hardest to implement, is to have the discipline to say no to deals that fall outside your core competence. The pressure to present a "well-diversified" portfolio to LPs is strong. But in a capital intensive era, diversification at the portfolio level is less valuable than conviction at the individual company level. The megafunds can afford diversification because their scale permits it. A small fund needs concentrated positions with deep diligence because its edge is not breadth of coverage but depth of insight. This is the "constructive compliance bridging" in the investment world: understanding which projects have real substance and which are merely decorated with a veneer of plausibility.
Let me now pivot to the policy dimension. The venture capital concentration has implications beyond the fund level. It affects which companies get built, which problems are solved, and which founders get a chance. In the past, the venture industry was a democratizing force in technology, funding a diverse range of ideas. The AI concentration has introduced a return to a more aristocratic structure where a small group of elites determines which technologies receive the oxygen of capital. We should not be naive about the consequences. When innovation is monopolized by a handful of gatekeepers, the social cost of a missed opportunity is enormous. The history of technology is full of examples where the consensus opinion was wrong and the marginal voice was right. The venture industry's breadth was its safety valve. The concentration we are observing now is thinning that valve.
Regulators, who are often several years behind the market, should be taking note. The venture industry has traditionally been exempt from most financial regulation, and for good reason. It is a private market with sophisticated investors. But the same regulatory frameworks that protect public investors are relevant to LPs who entrust their capital to external managers. The risk of concentration is systemic. If a handful of large funds make a coordinated bet on AI and that bet fails, the consequences will be felt not just by their LPs but by the startup ecosystem that depends on their continued investment. This is not a call for immediate intervention, but it is a warning that the incentives are becoming misaligned.
The crypto industry, which has its own checkered history with concentration, should be the first to recognize this alert. The centralization of power in any network creates a vulnerability manifold. The on-chain world has learned this through the collapse of FTX, the failure of Terra/LUNA, and the fragility of the centralized lending market in 2022. The venture industry is not a blockchain, but it operates on a similar principle. When too much power is concentrated in a small number of entities, the system becomes fragile. In a world facing geopolitical flux, energy transitions, and climate challenges, that fragility is the last thing we need.
I want to return, briefly, to the data reality that should be the ground truth for this article. The current market is one of concentration. The data tells us that the average valuation for AI startups in the seed stage has nearly doubled since 2022, while the valuation for non-AI SaaS startups has remained flat. The phenomenon is not a matter of opinion or PR spin. It is a measurable divergence. The funds participating in those top-of-market valuations are predominantly large funds, and the effect on small fund returns is structural. As AI deal sizes increase, the small fund's participation rate must necessarily decline or be spread across fewer positions. Either path forces a re-evaluation of the small fund's strategy.
For crypto funds, the warning is particularly stark. I now see crypto funds attempting to add "AI" to their thesis as a means of survival. Some do it because they genuinely understand the technical intersection — verifiable inference, decentralized rendering of training data, DAO-managed compute clusters. Most do it because they are afraid. They are mimicking the conversation because they fear being left out. This is the worst kind of strategy. It is not based on analysis; it is based on panic. The result will be a proliferation of diluted portfolios that satisfy neither the crypto world nor the AI world. The urgent need is for a clear-eyed analysis of what the actual opportunities are.
One such opportunity, which I have not yet seen adequately considered, is the combination of decentralized data provenance and AI governance. Large AI companies are facing mounting legal and regulatory pressure over their training data practices. This is not a fringe concern. It is a significant legal exposure that could reach billions of dollars. A solution that allows for transparent, immutable tracking of model training data — verifiability that goes beyond the "just trust us" model — would have enormous value. And that solution's architecture might look a lot like a public blockchain. The small fund that understands both AI governance and the limits of decentralized technology could build a position that no AI megafund can easily replicate, because the megafund's incentive structure is to keep the training data hidden.
That is, in a sense, the story of this entire article. The structural edge for small funds is not in trying to compete with the megafunds on their terms. It is in finding the spaces where the megafunds' incentives are misaligned with the market's actual needs, and then building a portfolio that serves those needs. The megafund wins when the frontier models succeed. The small fund wins when the output of those models is actually adopted in complex, real-world systems, which requires trust, verification, and domain-specific integration. Those are not problems that $1 billion can solve with a single check. They are solved by focused execution.
In the end, the division in venture capital is not simply about financial capacity. It is about what we, as an industry, are willing to count as intelligent risk-taking. The embrace of AI represents a bet that the current frontier technologies will realize their potential. Smaller funds being left behind, if they respond with panic and mimicry, will be a loss to the entire ecosystem. If they respond with discipline and focus, they may still find opportunities — not the same ones, but perhaps more durable ones.
Beauty is the mask; geometry is the bone. The beauty of the AI capital supercycle is that it appears to be a golden age of technological progress. The geometry is that it is also a structural redistribution of power. The funds that will survive are the ones that can see through the mask to the underlying structure, and the ones that will thrive are those that can find the cracks in that structure and build a position that is both intellectually and financially defensible.
As for the small funds that came of age in the last decade — the ones that built their identity on a particular sector or on regional agility — the road ahead is narrower, but it is not closed. The opportunity is not in owning the ore of the gold mine. It is in owning the picks and shovels, the equipment that makes the mining possible, and the maps that tell you where to dig. In the new AI economy, the small fund that owns the distribution channel, the domain expertise, and the verification tools is a player. The one that owns nothing but a hope to participate in the next mega-round is a bystander. We will need to decide which side we are on. The market will not decide for us.
The signposts are already visible. Follow the deal flow, not the press release. Measure the depth, not the wave. And remember that beneath the yield lies the rot, but beneath the rot, there is also fertile soil. The question is whether you have the patience to dig for it.